{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-08-11T09:21:12.743Z","headline":"Mistral 发布 Shieldstral：3B 开源权重多模态内容审核模型","description":"Mistral 推出 Shieldstral，一款 3B 参数的开源权重多模态内容审核模型。该模型面向文本与图像等内容的审核场景，权重开放可供开发者使用。","url":"https://www.aioga.com/news/cmsf5bjo21gmpro2e8p8xs5hu/","mainEntityOfPage":"https://www.aioga.com/news/cmsf5bjo21gmpro2e8p8xs5hu/","datePublished":"2026-08-04T20:41:07.887Z","dateModified":"2026-08-04T20:41:07.887Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://mistral.ai/news/shieldstral","https://aihot.virxact.com/items/cmsf5bjo21gmpro2e8p8xs5hu"],"canonicalUrl":"https://www.aioga.com/news/cmsf5bjo21gmpro2e8p8xs5hu/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：Mistral 推出 Shieldstral，一款 3B 参数的开源权重多模态内容审核模型。 Aioga 将其归入「模型更新」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/cmsf5bjo21gmpro2e8p8xs5hu/","dateCreated":"2026-08-04T20:41:07.887Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"mistral.ai source article","url":"https://mistral.ai/news/shieldstral","datePublished":"2026-08-04T20:41:07.887Z","provider":{"@type":"Organization","name":"mistral.ai","url":"https://mistral.ai/news/shieldstral"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmsf5bjo21gmpro2e8p8xs5hu","datePublished":"2026-08-04T20:41:07.887Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmsf5bjo21gmpro2e8p8xs5hu"}}],"aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","originalPublisher":{"name":"mistral.ai","url":"https://mistral.ai/news/shieldstral"},"geoDeepAnswer":null,"article":{"id":"cmsf5bjo21gmpro2e8p8xs5hu","slug":"cmsf5bjo21gmpro2e8p8xs5hu","url":"https://www.aioga.com/news/cmsf5bjo21gmpro2e8p8xs5hu/","title":"Mistral 发布 Shieldstral：3B 开源权重多模态内容审核模型","title_en":"Mistral 的 Shieldstral：适用于多模态内容审核的 3B 无权重限制模型","summary":"Mistral 推出 Shieldstral，一款 3B 参数的开源权重多模态内容审核模型。该模型面向文本与图像等内容的审核场景，权重开放可供开发者使用。","source":"Hacker News 热门（buzzing.cc 中文翻译）","sourceUrl":"https://mistral.ai/news/shieldstral","aiHotUrl":"https://aihot.virxact.com/items/cmsf5bjo21gmpro2e8p8xs5hu","publishedAt":"2026-08-04T20:41:07.887Z","category":"模型更新","score":47,"selected":false,"articleBody":["Build, test, and run AI agents and apps.","Train, align, and evaluate custom AI models.","Coding agents in the terminal, IDE, and background.","Frontier-scale infrastructure for training and inference.","Your Prompts and Skills need a system of record.","Shieldstral introduces a 3B open-weights multimodal safety classifier that outperforms models up to 7x its size by framing content moderation as a policy-adaptive question-answering task. Unlike traditional guardrail models, it accepts plain-language policies at inference time, unifying text and image safety evaluation without retraining. Released under Apache 2.0, it delivers calibrated safety scores across diverse benchmarks while running efficiently on a single 16GB NVIDIA GPU.","A 3B open-weights, policy-adaptive multimodal safety classifier that matches models up to 7x its size on text safety and sets a new state of the art on multimodal moderation.","“Does this content promote violence against a protected group? Is this image safe to show to a minor? Did the assistant refuse the request?”","Every product that ships a model needs to answer questions like these — but the right answer depends on the product, the audience, and the moment. The same content can be fine for a cybersecurity research tool and harmful on a mental-health platform. Most guardrail models bake a fixed taxonomy of harm categories into their weights, so re-targeting them to a new deployment context means retraining. And because safety definitions differ across applications and domains, there is no single \"correct\" set of categories to model in the first place.","Shieldstral takes a different approach: you write the policy as a plain-language question at inference time, and the model returns a calibrated safety score. No retraining, one interface for text and images, and a verdict from a single token. Please refer to our technical report：https://arxiv.org/abs/2607.25857 here.","As an inaugural member of the Open Secure AI Alliance：https://blogs.nvidia.com/blog/open-secure-ai-alliance/ with NVIDIA and other organizations, today we're releasing Shieldstral as open weights under Apache 2.0, available for download here：https://huggingface.co/mistralai/Shieldstral-1.0-3B.","Shieldstral frames content moderation as a binary question-answering task . Each request has three parts:","— the evaluation context, strictness, and (optionally) a definition of what counts as unsafe content.","— a single yes/no question, e.g. \"Does this content promote physical violence?\"","— the content to judge: a prompt, a response, a prompt–response pair, or an image with optional text.","At inference the model reads out only the yes and no logits and softmax-normalizes them into a continuous safety score. This one simple formulation does a lot of work: it unifies prompt classification, response moderation, refusal detection, and toxicity detection into a single problem; it lets policies live entirely in the prompt, so one checkpoint adapts to novel policies at deployment time.","Strong performance — matches or outperforms open guard models up to 7× its size across text safety, refusal detection, policy adaptability, and multimodal benchmarks.","Adaptive and flexible — a single natural-language interface covers text, image, and text+image content across prompts, responses, and prompt–response pairs. Policies are supplied as free-form queries and re-targeted at inference time, without retraining.","Small, trained on heterogeneous sources — a 3B model that runs on a single 16GB GPU, trained on real and synthetic data with diverse label formats and taxonomies, consolidated into one framework.","Continuous safety score — returns a calibrated yes / no probability from a single forward pass, so you can threshold or rank by confidence rather than relying on a discrete label.","We evaluate Shieldstral against open guard models up to 7x its size across four axes. All evaluation samples are held out from training.","The core idea is that a small model can beat much larger ones if the data is right. Getting the data right meant solving four problems:","Unify heterogeneous data. Public safety datasets disagree on taxonomies, labels, and annotation conventions — from binary safe/unsafe flags to fine-grained multi-label taxonomies. We convert every dataset into the same instruction–query–document format with a per-dataset processor, and we vary the wording of instructions, queries, and prompt–response delimiters so the model generalizes across phrasing instead of overfitting to one style. We also calibrate strictness per source — strict for adversarial jailbreaks, lenient for response-quality data — so the model learns calibrated decision boundaries. This lets us consolidate sources that would otherwise be incompatible.","Teach discrimination, not memorization. If trained on a fixed set of policy labels, a model learns only to classify those predefined policies, rather than reasoning about the precise boundaries of a given policy. This prevents generalization to novel policies. Instead, we construct sets of deliberately similar, easily confused policies and ask an LLM to rewrite safe text into contrastive pairs: each rewrite is engineered to violate one policy but not its sibling. This trains the model to distinguish which specific policy a piece of content violates , a skill that transfers to unseen, user-defined policies at inference time.","Ground safety in images. Unsafe images can't be synthezised by an LLM the way text can, so visual safety data is scarce. We supplement limited moderation datasets with general-purpose image datasets as high-quality negatives, mutate queries to augment the dataset, and filter every image–query pair through a vision–language reranker to reduce mislabeled data and hallucinations.","Combine complementary checkpoints. We fine-tune with LoRA and merge — via SLERP — a checkpoint calibrated on public data, one that adds fine-grained policy discrimination from generated data, and the base instruct model. The merge recovers common policy calibration and policy adaptability in a single model, and instruction-following from the base model transfers to the moderation task.","Forge . We built Shieldstral end to end on Forge：https://mistral.ai/products/forge/, our platform for training, aligning, and evaluating custom models. Forge managed the infrastructure, data and model sharding, metrics, and logging on top of state-of-the-art distributed training, so the team could stay focused on the data which is what determines the safety model's quality.","Shieldstral is a step toward moderation that adapts to context instead of forcing every product through one frozen taxonomy. We're continuing to push on multilingual coverage, longer-document robustness, and broader multimodal safety — and we'd love to see what the community builds on top of it.","BTW, we're hiring! If you want to help make AI better, see our careers page ：https://mistral.ai/careers ."],"articleImages":[{"sourceUrl":"https://mistral.ai/cms-media/api/media/file/2a1ffaf3-f171-460b-be1b-fc734aa776aa.svg","alt":"2a1ffaf3-f171-460b-be1b-fc734aa776aa","afterParagraph":3,"url":"/media/articles/cmses2cpo14kuro2ehdj6irxa/e32e6b666122f9f7.jpg"},{"sourceUrl":"https://mistral.ai/cms-media/api/media/file/icon-m-microphone.svg","alt":"","afterParagraph":3,"url":"/media/articles/cmses2cpo14kuro2ehdj6irxa/c42626ebf983a308.jpg"}],"mediaStatus":"ok","articleBodyZh":["构建、测试并运行 AI 代理和应用程序。","训练、校准并评估自定义 AI 模型。","在终端、IDE 和后台编写代理代码。","用于训练和推理的前沿规模基础设施。","你的提示和技能需要一个记录系统。","Shieldstral 推出了一个 30 亿参数开源权重多模态安全分类器，通过将内容审核框架化为政策自适应问答任务，其性能超过大约 7 倍规模的模型。与传统的护栏模型不同，它在推理时接受自然语言政策，无需重新训练即可统一文本和图像的安全评估。此模型在 Apache 2.0 许可证下发布，在运行于单个 16GB NVIDIA GPU 时，可在各种基准上提供校准的安全评分并高效运行。","一款 30 亿参数、政策自适应的开源权重多模态安全分类器，在文本安全方面与大约 7 倍自身规模的模型表现相当，并在多模态审核上创下新的技术标准。","“该内容是否鼓励对受保护群体的暴力？该图像是否适合未成年人观看？助手是否拒绝了请求？”","每款发布模型的产品都需要回答类似问题——但正确答案取决于产品、受众和具体时刻。同一内容在网络安全研究工具上可能无害，而在心理健康平台上可能有害。大多数护栏模型在权重中内置了固定的危害类别，因此将它们重新定位到新部署环境意味着必须重新训练。而且由于安全定义在应用和领域中有所不同，根本不存在唯一“正确”的类别集合用于建模。","Shieldstral 采用了不同的方法：你可以在推理时将政策写成自然语言问题，模型返回校准后的安全评分。无需重新训练，文本和图像共用一个接口，单个 token 即可得出裁定。请参阅我们的技术报告：https://arxiv.org/abs/2607.25857。","作为 Open Secure AI Alliance 的创始成员：https://blogs.nvidia.com/blog/open-secure-ai-alliance/，与 NVIDIA 及其他组织合作，今天我们以 Apache 2.0 许可证开放发布 Shieldstral 权重，可在此下载：https://huggingface.co/mistralai/Shieldstral-1.0-3B。","Shieldstral 将内容审核框架设定为一个二元问答任务。每个请求包含三部分：","— 评估上下文、严格程度，以及（可选的）不安全内容定义。","— 一个是/否问题，例如：“此内容是否促进身体暴力？”","— 待判断内容：一个提示、一个回复、一个提示—回复对，或带可选文本的图像。","在推理时，模型只读取是与否的对数几率，并将其通过 softmax 转换为连续的安全评分。这个简单的公式做了很多工作：它将提示分类、回复审核、拒绝检测和有害内容检测统一为一个问题；它使策略完全存在于提示中，因此一个检查点在部署时即可适应新的策略。","强大性能 — 在文本安全、拒绝检测、策略适应性和多模态基准测试中，性能匹配或超过其 7 倍规模的开源守护模型。","自适应且灵活 — 单一自然语言接口涵盖提示、回复以及提示—回复对的文本、图像和文本+图像内容。策略以自由形式查询提供，并在推理时重新定位，无需重新训练。","小型，异质数据训练 — 一个 3B 模型，可在单个 16GB GPU 上运行，训练数据包含真实和合成数据，且标签格式与分类系统多样化，并整合在一个框架中。","连续安全评分 — 通过单次前向传播返回校准后的是/否概率，因此您可以根据信心进行阈值或排序，而不是依赖离散标签。","我们在四个维度上对 Shieldstral 进行了评估，其规模最高达开源守护模型的 7 倍。所有评估样本均未用于训练。","核心思想是，如果数据正确，小模型可以打败大得多的模型。把数据做对意味着解决四个问题：","统一异构数据。公共安全数据集在分类、标签和标注规范上存在差异——从二元的安全/不安全标记到细粒度的多标签分类。我们使用每个数据集的处理器将每个数据集转换为相同的指令-查询-文档格式，并且我们会变化指令、查询以及提示-响应分隔符的措辞，使模型能够在不同措辞之间泛化，而不会过拟合于某一种风格。我们还根据来源调整严格性——针对对抗性越狱严格，对响应质量数据宽松——使模型学习校准的决策边界。这使得我们能够整合原本不兼容的数据源。","教会区分，而不是记忆。如果仅训练于固定的政策标签集合，模型只会学会对这些预定义政策进行分类，而无法推理出某一政策的精确界限。这会阻碍对新政策的泛化。相反，我们构建成对相似、容易混淆的政策集，并要求大型语言模型将安全文本重写为对比对：每个重写都旨在违反一个政策而不违反其类似政策。这训练模型识别内容违反的是具体哪一条政策，这一能力可以在推理时迁移到未见过的、用户定义的政策上。","以图像为基础进行安全控制。与文本不同，不安全图像无法由大型语言模型生成，因此视觉安全数据稀缺。我们利用通用图像数据集作为高质量的负样本，补充有限的审核数据集，对查询进行变异以扩充数据集，并通过视觉-语言重排序器过滤每个图像-查询对，以减少错误标注数据和幻觉。","组合互补的检查点。我们使用LoRA进行微调，并通过SLERP合并——一个在公开数据上校准的检查点、一个从生成数据中增加细粒度政策区分的检查点，以及基础指令模型。该合并在单一模型中恢复了常见的政策校准和政策适应性，同时基础模型的指令执行能力可以迁移到审核任务上。","Forge。我们在 Forge 上端到端构建了 Shieldstral：https://mistral.ai/products/forge/，这是我们用于训练、校准和评估自定义模型的平台。Forge 管理基础设施、数据和模型分片、指标和日志记录，并基于最先进的分布式训练技术，因此团队可以专注于数据，而数据决定了安全模型的质量。","Shieldstral 是向适应上下文的内容审核迈出的一步，而不是将每个产品强行套入一个固定的分类体系。我们正在继续推进多语言覆盖、更长文档的稳健性以及更广泛的多模态安全——我们也非常期待看到社区在其基础上构建的成果。","顺便说一下，我们正在招聘！如果你想帮助改善 AI，请查看我们的招聘页面：https://mistral.ai/careers。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：Mistral 推出 Shieldstral，一款 3B 参数的开源权重多模态内容审核模型。 Aioga 将其归入「模型更新」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：模型与研究类动态需要结合能力边界、开放方式、成本、可用性和真实任务表现判断，单项指标领先不等于已经形成稳定采用。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察官方文档、实际可用性、价格变化、开发者反馈和竞品回应。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-08-11T09:23:27.680Z","sourceHash":"3b7f458f60278742","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["模型更新","Hacker News 热门（buzzing.cc 中文翻译）"],"translations":{"zh-CN":{"title":"Introducing Shieldstral. August 4， 2026 By Mistral","summary":"Shieldstral introduces a 3B open-weights multimodal safety classifier that outperforms models up to 7x its size.","category":"AI资讯","source":"Mistral AI：News（网页）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Introducing Shieldstral. August 4， 2026 By Mistral - Aioga AI资讯","description":"Shieldstral introduces a 3B open-weights multimodal safety classifier that outperforms models up to 7x its size.","url":"https://www.aioga.com/news/cmsf5bjo21gmpro2e8p8xs5hu/","articleBody":["构建、测试并运行 AI 代理和应用程序。","训练、校准并评估自定义 AI 模型。","在终端、IDE 和后台编写代理代码。","用于训练和推理的前沿规模基础设施。","你的提示和技能需要一个记录系统。","Shieldstral 推出了一个 30 亿参数开源权重多模态安全分类器，通过将内容审核框架化为政策自适应问答任务，其性能超过大约 7 倍规模的模型。与传统的护栏模型不同，它在推理时接受自然语言政策，无需重新训练即可统一文本和图像的安全评估。此模型在 Apache 2.0 许可证下发布，在运行于单个 16GB NVIDIA GPU 时，可在各种基准上提供校准的安全评分并高效运行。","一款 30 亿参数、政策自适应的开源权重多模态安全分类器，在文本安全方面与大约 7 倍自身规模的模型表现相当，并在多模态审核上创下新的技术标准。","“该内容是否鼓励对受保护群体的暴力？该图像是否适合未成年人观看？助手是否拒绝了请求？”","每款发布模型的产品都需要回答类似问题——但正确答案取决于产品、受众和具体时刻。同一内容在网络安全研究工具上可能无害，而在心理健康平台上可能有害。大多数护栏模型在权重中内置了固定的危害类别，因此将它们重新定位到新部署环境意味着必须重新训练。而且由于安全定义在应用和领域中有所不同，根本不存在唯一“正确”的类别集合用于建模。","Shieldstral 采用了不同的方法：你可以在推理时将政策写成自然语言问题，模型返回校准后的安全评分。无需重新训练，文本和图像共用一个接口，单个 token 即可得出裁定。请参阅我们的技术报告：https://arxiv.org/abs/2607.25857。","作为 Open Secure AI Alliance 的创始成员：https://blogs.nvidia.com/blog/open-secure-ai-alliance/，与 NVIDIA 及其他组织合作，今天我们以 Apache 2.0 许可证开放发布 Shieldstral 权重，可在此下载：https://huggingface.co/mistralai/Shieldstral-1.0-3B。","Shieldstral 将内容审核框架设定为一个二元问答任务。每个请求包含三部分：","— 评估上下文、严格程度，以及（可选的）不安全内容定义。","— 一个是/否问题，例如：“此内容是否促进身体暴力？”","— 待判断内容：一个提示、一个回复、一个提示—回复对，或带可选文本的图像。","在推理时，模型只读取是与否的对数几率，并将其通过 softmax 转换为连续的安全评分。这个简单的公式做了很多工作：它将提示分类、回复审核、拒绝检测和有害内容检测统一为一个问题；它使策略完全存在于提示中，因此一个检查点在部署时即可适应新的策略。","强大性能 — 在文本安全、拒绝检测、策略适应性和多模态基准测试中，性能匹配或超过其 7 倍规模的开源守护模型。","自适应且灵活 — 单一自然语言接口涵盖提示、回复以及提示—回复对的文本、图像和文本+图像内容。策略以自由形式查询提供，并在推理时重新定位，无需重新训练。","小型，异质数据训练 — 一个 3B 模型，可在单个 16GB GPU 上运行，训练数据包含真实和合成数据，且标签格式与分类系统多样化，并整合在一个框架中。","连续安全评分 — 通过单次前向传播返回校准后的是/否概率，因此您可以根据信心进行阈值或排序，而不是依赖离散标签。","我们在四个维度上对 Shieldstral 进行了评估，其规模最高达开源守护模型的 7 倍。所有评估样本均未用于训练。","核心思想是，如果数据正确，小模型可以打败大得多的模型。把数据做对意味着解决四个问题：","统一异构数据。公共安全数据集在分类、标签和标注规范上存在差异——从二元的安全/不安全标记到细粒度的多标签分类。我们使用每个数据集的处理器将每个数据集转换为相同的指令-查询-文档格式，并且我们会变化指令、查询以及提示-响应分隔符的措辞，使模型能够在不同措辞之间泛化，而不会过拟合于某一种风格。我们还根据来源调整严格性——针对对抗性越狱严格，对响应质量数据宽松——使模型学习校准的决策边界。这使得我们能够整合原本不兼容的数据源。","教会区分，而不是记忆。如果仅训练于固定的政策标签集合，模型只会学会对这些预定义政策进行分类，而无法推理出某一政策的精确界限。这会阻碍对新政策的泛化。相反，我们构建成对相似、容易混淆的政策集，并要求大型语言模型将安全文本重写为对比对：每个重写都旨在违反一个政策而不违反其类似政策。这训练模型识别内容违反的是具体哪一条政策，这一能力可以在推理时迁移到未见过的、用户定义的政策上。","以图像为基础进行安全控制。与文本不同，不安全图像无法由大型语言模型生成，因此视觉安全数据稀缺。我们利用通用图像数据集作为高质量的负样本，补充有限的审核数据集，对查询进行变异以扩充数据集，并通过视觉-语言重排序器过滤每个图像-查询对，以减少错误标注数据和幻觉。","组合互补的检查点。我们使用LoRA进行微调，并通过SLERP合并——一个在公开数据上校准的检查点、一个从生成数据中增加细粒度政策区分的检查点，以及基础指令模型。该合并在单一模型中恢复了常见的政策校准和政策适应性，同时基础模型的指令执行能力可以迁移到审核任务上。","Forge。我们在 Forge 上端到端构建了 Shieldstral：https://mistral.ai/products/forge/，这是我们用于训练、校准和评估自定义模型的平台。Forge 管理基础设施、数据和模型分片、指标和日志记录，并基于最先进的分布式训练技术，因此团队可以专注于数据，而数据决定了安全模型的质量。","Shieldstral 是向适应上下文的内容审核迈出的一步，而不是将每个产品强行套入一个固定的分类体系。我们正在继续推进多语言覆盖、更长文档的稳健性以及更广泛的多模态安全——我们也非常期待看到社区在其基础上构建的成果。","顺便说一下，我们正在招聘！如果你想帮助改善 AI，请查看我们的招聘页面：https://mistral.ai/careers。"]},"en":{"title":"Mistral releases Shieldstral: 3B open-source weighted multimodal content moderation model","summary":"Mistral launched Shieldstral, a 3B-parameter open-source weighted multimodal content moderation model. This model is designed for review scenarios involving text and images, with open weights available for developers to use.","category":"Models","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral releases Shieldstral: 3B open-source weighted multimodal content moderation model - Aioga AI News","description":"Mistral launched Shieldstral, a 3B-parameter open-source weighted multimodal content moderation model. This model is designed for review scenarios involving text and images, with o...","url":"https://www.aioga.com/en/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:23.005Z"},"ja":{"title":"MistralがShieldstralをリリース:3Bオープンソースの重み付きマルチモーダルコンテンツモデレーションモデル","summary":"Mistralは、3Bパラメータのオープンソース加重マルチモーダルコンテンツモデレーションモデルであるShieldstralを立ち上げました。 このモデルはテキストや画像を含むレビューシナリオ向けに設計されており、開発者が利用できるオープンウェイトが用意されています。","category":"モデル更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"MistralがShieldstralをリリース:3Bオープンソースの重み付きマルチモーダルコンテンツモデレーションモデル - Aioga AIニュース","description":"Mistralは、3Bパラメータのオープンソース加重マルチモーダルコンテンツモデレーションモデルであるShieldstralを立ち上げました。 このモデルはテキストや画像を含むレビューシナリオ向けに設計されており、開発者が利用できるオープンウェイトが用意されています。","url":"https://www.aioga.com/ja/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:23.628Z"},"ko":{"title":"Mistral, Shieldstral을 출시했습니다: 3B 오픈소스 가중 멀티모달 콘텐츠 관리 모델","summary":"Mistral은 3B 매개변수 오픈소스 가중 멀티모달 콘텐츠 관리 모델인 Shieldstral을 출시했습니다. 이 모델은 텍스트와 이미지가 포함된 검토 시나리오를 위해 설계되었으며, 개발자가 사용할 수 있는 오픈 가중치가 제공됩니다.","category":"모델 업데이트","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral, Shieldstral을 출시했습니다: 3B 오픈소스 가중 멀티모달 콘텐츠 관리 모델 - Aioga AI 뉴스","description":"Mistral은 3B 매개변수 오픈소스 가중 멀티모달 콘텐츠 관리 모델인 Shieldstral을 출시했습니다. 이 모델은 텍스트와 이미지가 포함된 검토 시나리오를 위해 설계되었으며, 개발자가 사용할 수 있는 오픈 가중치가 제공됩니다.","url":"https://www.aioga.com/ko/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:32.431Z"},"es":{"title":"Mistral lanza Shieldstral: modelo de moderación de contenido multimodal ponderado de código abierto 3B","summary":"Mistral lanzó Shieldstral, un modelo de moderación de contenido multimodal ponderado de código abierto y código abierto. Este modelo está diseñado para escenarios de revisión que involucran texto e imágenes, con pesos abiertos disponibles para que los desarrolladores los utilicen.","category":"Modelos","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral lanza Shieldstral: modelo de moderación de contenido multimodal ponderado de código abierto 3B - Aioga Noticias de IA","description":"Mistral lanzó Shieldstral, un modelo de moderación de contenido multimodal ponderado de código abierto y código abierto. Este modelo está diseñado para escenarios de revisión que i...","url":"https://www.aioga.com/es/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:32.496Z"},"fr":{"title":"Mistral publie Shieldstral : modèle de modération de contenu multimodal pondé open-source 3B","summary":"Mistral a lancé Shieldstral, un modèle de modération de contenu multimodal pondéré à 3B paramètres open source. Ce modèle est conçu pour des scénarios de révision impliquant texte et images, avec des poids ouverts disponibles pour les développeurs.","category":"Modèles","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral publie Shieldstral : modèle de modération de contenu multimodal pondé open-source 3B - Aioga Actualités IA","description":"Mistral a lancé Shieldstral, un modèle de modération de contenu multimodal pondéré à 3B paramètres open source. Ce modèle est conçu pour des scénarios de révision impliquant texte...","url":"https://www.aioga.com/fr/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:41.328Z"},"de":{"title":"Mistral veröffentlicht das Open-Source-Modell der gewichteten multimodalen Inhaltsmoderation Shieldstral: 3B","summary":"Mistral brachte Shieldstral auf den Markt, ein 3B-Parameter Open-Source-Weighted-Modell zur multimodalen Inhaltsmoderation. Dieses Modell ist für Wiederholungsszenarien mit Text und Bildern konzipiert, wobei offene Gewichte für Entwickler zur Verfügung stehen.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral veröffentlicht das Open-Source-Modell der gewichteten multimodalen Inhaltsmoderation Shieldstral: 3B - Aioga KI-News","description":"Mistral brachte Shieldstral auf den Markt, ein 3B-Parameter Open-Source-Weighted-Modell zur multimodalen Inhaltsmoderation. Dieses Modell ist für Wiederholungsszenarien mit Text un...","url":"https://www.aioga.com/de/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:41.251Z"},"pt-BR":{"title":"Mistral lança Shieldstral: modelo de moderação de conteúdo multimodal ponderado open-source 3B","summary":"A Mistral lançou o Shieldstral, um modelo de moderação de conteúdo multimodal ponderado e open-source de código aberto e de 3B. Este modelo foi projetado para cenários de revisão envolvendo texto e imagens, com pesos abertos disponíveis para desenvolvedores usarem.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral lança Shieldstral: modelo de moderação de conteúdo multimodal ponderado open-source 3B - Aioga Notícias de IA","description":"A Mistral lançou o Shieldstral, um modelo de moderação de conteúdo multimodal ponderado e open-source de código aberto e de 3B. Este modelo foi projetado para cenários de revisão e...","url":"https://www.aioga.com/pt-BR/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:50.138Z"},"ru":{"title":"Mistral выпускает Shieldstral: 3B открытую модель взвешенной мультимодальной модерации контента","summary":"Mistral запустила Shieldstral — модель модерации мультимодального контента с 3B-параметрами с открытым исходным кодом. Эта модель предназначена для обзорных сценариев с текстом и изображениями, с доступными открытыми весами для разработчиков.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral выпускает Shieldstral: 3B открытую модель взвешенной мультимодальной модерации контента - Aioga Новости ИИ","description":"Mistral запустила Shieldstral — модель модерации мультимодального контента с 3B-параметрами с открытым исходным кодом. Эта модель предназначена для обзорных сценариев с текстом и и...","url":"https://www.aioga.com/ru/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:50.132Z"},"ar":{"title":"تطلق ميسترال نموذج Shieldstral: 3B مفتوح المصدر وموزون لمراقبة المحتوى متعدد الوسائط","summary":"أطلقت ميسترال شيلدسترال، وهو نموذج إشراف محتوى متعدد الوسائط ومفتوح المصدر متعدد الوسائط بمعيار 3B. تم تصميم هذا النموذج لسيناريوهات المراجعة التي تتضمن نصا وصورا، مع أوزان مفتوحة متاحة للمطورين لاستخدامها.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"تطلق ميسترال نموذج Shieldstral: 3B مفتوح المصدر وموزون لمراقبة المحتوى متعدد الوسائط - Aioga أخبار الذكاء الاصطناعي","description":"أطلقت ميسترال شيلدسترال، وهو نموذج إشراف محتوى متعدد الوسائط ومفتوح المصدر متعدد الوسائط بمعيار 3B. تم تصميم هذا النموذج لسيناريوهات المراجعة التي تتضمن نصا وصورا، مع أوزان مفتوحة...","url":"https://www.aioga.com/ar/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:58.870Z"},"hi":{"title":"मिस्ट्रल ने शील्डस्ट्रल जारी किया: 3बी ओपन-सोर्स भारित मल्टीमॉडल सामग्री मॉडरेशन मॉडल","summary":"मिस्ट्रल ने शील्डस्ट्रल लॉन्च किया, जो एक 3बी-पैरामीटर ओपन-सोर्स वेटेड मल्टीमॉडल कंटेंट मॉडरेशन मॉडल है। यह मॉडल पाठ और छवियों से जुड़े समीक्षा परिदृश्यों के लिए डिज़ाइन किया गया है, जिसमें डेवलपर्स के उपयोग के लिए खुले वजन उपलब्ध हैं।","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"मिस्ट्रल ने शील्डस्ट्रल जारी किया: 3बी ओपन-सोर्स भारित मल्टीमॉडल सामग्री मॉडरेशन मॉडल - Aioga AI समाचार","description":"मिस्ट्रल ने शील्डस्ट्रल लॉन्च किया, जो एक 3बी-पैरामीटर ओपन-सोर्स वेटेड मल्टीमॉडल कंटेंट मॉडरेशन मॉडल है। यह मॉडल पाठ और छवियों से जुड़े समीक्षा परिदृश्यों के लिए डिज़ाइन किया गया ह...","url":"https://www.aioga.com/hi/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:21:58.875Z"},"it":{"title":"Mistral rilascia Shieldstral: modello di moderazione multimodale open-source ponderata 3B","summary":"Mistral ha lanciato Shieldstral, un modello di moderazione dei contenuti multimodali ponderati open-source a 3B. Questo modello è progettato per scenari di revisione che coinvolgono testo e immagini, con pesi a vuoto disponibili per gli sviluppatori.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral rilascia Shieldstral: modello di moderazione multimodale open-source ponderata 3B - Aioga Notizie IA","description":"Mistral ha lanciato Shieldstral, un modello di moderazione dei contenuti multimodali ponderati open-source a 3B. Questo modello è progettato per scenari di revisione che coinvolgon...","url":"https://www.aioga.com/it/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:08.217Z"},"nl":{"title":"Mistral brengt Shieldstral: 3B open-source gewogen multimodale contentmoderatiemodel uit","summary":"Mistral lanceerde Shieldstral, een 3B-parameter open-source gewogen multimodale contentmoderatiemodel. Dit model is ontworpen voor reviewscenario's met tekst en afbeeldingen, met open gewichten beschikbaar voor ontwikkelaars.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral brengt Shieldstral: 3B open-source gewogen multimodale contentmoderatiemodel uit - Aioga AI-nieuws","description":"Mistral lanceerde Shieldstral, een 3B-parameter open-source gewogen multimodale contentmoderatiemodel. Dit model is ontworpen voor reviewscenario's met tekst en afbeeldingen, met o...","url":"https://www.aioga.com/nl/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:07.273Z"},"tr":{"title":"Mistral, Shieldstral'ı yayınladı: 3B açık kaynaklı ağırlıklı çok modlu içerik moderasyon modeli","summary":"Mistral, 3B parametreli açık kaynaklı ağırlıklı multimodal içerik moderasyon modeli olan Shieldstral'ı başlattı. Bu model, metin ve görselleri içeren senaryoları incelemek için tasarlanmıştır ve geliştiricilerin kullanabileceği açık ağırlıklar mevcuttur.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral, Shieldstral'ı yayınladı: 3B açık kaynaklı ağırlıklı çok modlu içerik moderasyon modeli - Aioga AI Haberleri","description":"Mistral, 3B parametreli açık kaynaklı ağırlıklı multimodal içerik moderasyon modeli olan Shieldstral'ı başlattı. Bu model, metin ve görselleri içeren senaryoları incelemek için tas...","url":"https://www.aioga.com/tr/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:16.599Z"},"vi":{"title":"Mistral phát hành mô hình kiểm duyệt nội dung đa phương thức mã nguồn mở Shieldstral: 3B","summary":"Mistral đã ra mắt Shieldstral, một mô hình kiểm duyệt nội dung đa phương thức mã nguồn mở có trọng số 3B. Mô hình này được thiết kế cho các kịch bản đánh giá liên quan đến văn bản và hình ảnh, với trọng số mở có sẵn để các nhà phát triển sử dụng.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral phát hành mô hình kiểm duyệt nội dung đa phương thức mã nguồn mở Shieldstral: 3B - Tin tức AI Aioga","description":"Mistral đã ra mắt Shieldstral, một mô hình kiểm duyệt nội dung đa phương thức mã nguồn mở có trọng số 3B. Mô hình này được thiết kế cho các kịch bản đánh giá liên quan đến văn bản...","url":"https://www.aioga.com/vi/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:17.159Z"},"id":{"title":"Mistral merilis Shieldstral: model moderasi konten multimodal berbobot 3B open-source","summary":"Mistral meluncurkan Shieldstral, model moderasi konten multimodal berbobot open-source 3B-parameter. Model ini dirancang untuk skenario tinjauan yang melibatkan teks dan gambar, dengan bobot terbuka tersedia untuk digunakan oleh pengembang.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral merilis Shieldstral: model moderasi konten multimodal berbobot 3B open-source - Berita AI Aioga","description":"Mistral meluncurkan Shieldstral, model moderasi konten multimodal berbobot open-source 3B-parameter. Model ini dirancang untuk skenario tinjauan yang melibatkan teks dan gambar, de...","url":"https://www.aioga.com/id/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:25.892Z"},"th":{"title":"Mistral เปิดตัวโมเดลการควบคุมเนื้อหาแบบหลายรูปแบบแบบโอเพ่นซอร์ส Shieldstral: 3B แบบโอเพ่นซอร์ส","summary":"Mistral เปิดตัว Shieldstral ซึ่งเป็นโมเดลการควบคุมเนื้อหาแบบหลายรูปแบบแบบแบบโอเพ่นซอร์สที่มีพารามิเตอร์ 3B โมเดลนี้ออกแบบมาสําหรับสถานการณ์ทบทวนที่เกี่ยวข้องกับข้อความและภาพ โดยมีน้ําหนักเปิดให้นักพัฒนาใช้","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral เปิดตัวโมเดลการควบคุมเนื้อหาแบบหลายรูปแบบแบบโอเพ่นซอร์ส Shieldstral: 3B แบบโอเพ่นซอร์ส - ข่าว AI Aioga","description":"Mistral เปิดตัว Shieldstral ซึ่งเป็นโมเดลการควบคุมเนื้อหาแบบหลายรูปแบบแบบแบบโอเพ่นซอร์สที่มีพารามิเตอร์ 3B โมเดลนี้ออกแบบมาสําหรับสถานการณ์ทบทวนที่เกี่ยวข้องกับข้อความและภาพ โดยมีน...","url":"https://www.aioga.com/th/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:25.849Z"},"pl":{"title":"Mistral udostępnia Shieldstral: 3B open-source model ważonej, multimodalnej moderacji treści","summary":"Mistral uruchomił Shieldstral, 3B-parametrowy, otwartoźródłowy, ważony multimodalny model moderacji treści. Model ten został zaprojektowany do przeglądu scenariuszy z udziałem tekstu i obrazów, z otwartymi wagami dostępnymi dla deweloperów.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Mistral AI：News（网页）","pageTitle":"Mistral udostępnia Shieldstral: 3B open-source model ważonej, multimodalnej moderacji treści - Aioga Wiadomości AI","description":"Mistral uruchomił Shieldstral, 3B-parametrowy, otwartoźródłowy, ważony multimodalny model moderacji treści. Model ten został zaprojektowany do przeglądu scenariuszy z udziałem teks...","url":"https://www.aioga.com/pl/news/cmsf5bjo21gmpro2e8p8xs5hu/","contentTranslated":true,"sourceHash":"29f969723268e072","translatedAt":"2026-08-04T21:22:34.750Z"}}}}